Přístupy strojového učení v úloze klasifikace rentgenových snímků

The diploma thesis discusses two key machine learning approaches used to solve the classification task over images. It aims to verify, by performing a comparative classification task over lung X-ray images, whether models based on an approach using CNN method show better classification performance in a particular task than models based on a traditional approach combining unsupervised feature extraction and the use of traditional machine learning methods. The two approaches, whose identification is the result of a description of the genesis of the Computer Vision discipline and a search mapping current trends in the field of automatic detection of pneumonia from lung X-ray images, are contrasted within a specific classification task. The evaluation of the model sets from both approaches demonstrates the dominance of models based on the CNN approach over those based on the traditional approach, thus essentially confirming in a partial way the role of the CNN method as a state-of-the-art approach in the field of supervised image classification.